stevensonetal2000

8/14/2019 stevensonetal2000

1/38

Interpreting pronouns and connectives: Interactions

among focusing, thematic roles and coherence

relations

Rosemary StevensonUniversity of Durham, Durham, UK

Alistair Knott and Jon OberlanderUniversity of Edinburgh, Edinburgh, UK

Sharon McDonaldUniversity of Durham, Durham, UK

This paper investigates the relationship between focusing and coherencerelations in pronoun comprehension. In their focusing model of pronouncomprehension, Stevenson, Crawley and Kleinman (1994) proposed a defaultfocus on the thematic role associated with the consequences of a describedevent, a focus that may be modied by the attention-directing properties of asubsequent connective. In this paper we examine a second function ofconnectives: that of signalling the coherence relations between two clauses(e.g., a NARRATIVE relation or a RESULT relation). In three studies, weidentied the coherence relations between sentence fragments ending inpronouns and participants continuations to the fragments. We thenexamined the relationship between the coherence relation, the preferredreferent of the pronoun and the referents thematic role. The results ofstudies 1 and 2 showed that people aim to keep the focused entity, the

Requests for reprints should be addressed to Jon Oberlander, Human CommunicationResearch Centre, Division of Informatics, University of Edinburgh, 2 Buccleuch Place,Edinburgh, EH8 9LW, UK. E-mail: [email protected]

The support of the Economic and Social Research Council for the Human Communication

Research Centre is gratefully acknowledged. The second author was supported by theIntelligent Labelling Explorer project, funded by the Engineering and Physical SciencesResearch Council, under grant GR/K53321. The third author was supported by an EPSRCAdvanced Fellowship. Our thanks to Massimo Poesio for valuable discussions on the issuesinvestigated in this paper. Many thanks also to David Kleinman for helping to make the paper

readable. We also gratefully acknowledge the assistance of three anonymous reviewers forcomments on a draft of this paper.

c 2000 Psychology Press Ltd

http://www.tandf.co.uk/journals/pp/01690965.html

LANGUAGE AND COGNITIVE PROCESSES, 2000, 15 (3), 225262


2/38

226 STEVENSON ET AL.

coherence relation and the referent of the pronoun in alignment. Study 3included the connective next, which enabled us to generate differentpredictions for the roles of focusing and coherence relations in pronounresolution. The results favoured the focusing view. The preferred referent ofthe pronoun was the focused, rst mentioned, individual, whereas the

coherence relation was consistent with the thematic role of the pronominalreferent. If the pronoun referred to an Agent, a NARRATIVE relation waspreferred, if the pronoun referred to a Patient, a RESULT relation waspreferred. Discussion of these and other results led to the followingconclusions. First, pronoun resolution is primarily determined by focusing,either semantic or structural, although a range of other features, includingcoherence relations and verb semantics, may also act as pressures on pronounresolution. Second, the consistent link we observed between thematic rolesand coherence relations may provide a mapping between a represented

entity and a represented event. Third, the connectives we used have threedistinct functions: an attention directing function, a function for constrainingthe possible coherence relation between two events, and a function forinterpreting a clause as having either a causal or a temporal structure.

INTRODUCTION

A recurrent nding in research on the comprehension of pronouns is thatsome antecedents are more accessible than others. The prevalentexplanation of such effects is that features of the discourse direct thecomprehenders attention to a specic element in the discourse model.This element then becomes the focus of the comprehenders attention sothat when a pronoun is encountered, the pronoun is interpreted asreferring to this most highly focused element. As long as the interpretationis compatible with subsequent content, then comprehension ows

smoothly; if the interpretation is incompatible with subsequent content,then comprehension is impaired because an alternative referent for thepronoun has to be found. According to this view, pronoun comprehensionis primarily a top-down process, and the most focused entity in thediscourse is assumed to be the pronouns referent. We refer to thisexplanation as the focusing hypothesis. An alternative possibility is thatthe accessibility of a referent for a pronoun is inuenced not by focusingbut by the relationship between the events described in the discourse(Hobbs, 1979). That is, when comprehending a pronoun, in a sentence suchas John telephoned Bill because he needed some information, thecomprehender rst identies the relation between the two clauses andthen interprets the pronoun as referring to the individual most likely to beinvolved in that relation. We refer to this explanation as the relationalhypothesis. The relational hypothesis states that when two referents areintroduced in one clause and a pronoun refers to one of them in a second

clause, then the perceived relationship between the two described events


3/38

INTERPRETING PRONOUNS AND CONNECTIVES 227

determines the interpretation of the pronoun. The aim of our paper is toexamine these two hypotheses.

In what follows, we rst review the focusing hypothesis, according towhich connectives have focusing properties, and we show how focusing

explains pronoun resolution. We then review the relational hypothesis,according to which the connective identies the coherence relationbetween two clauses, and we show how the perceived coherence relationcan explain pronoun resolution. Then we place our work in the context ofother research on discourse and pronoun resolution, after which we outlinethe methods we use and spell out our predictions according to eachhypothesis. We then report three studies and discuss their results in thelight of our hypotheses.

The focusing hypothesis

Focusing models differ in what they regard as the underlying mechanism.Some researchers have argued for structural focusing (Grosz, Joshi, &Weinstein, 1983), whereas others have argued for semantic focusing(Stevenson et al., 1994) or focusing based on background knowledge of the

topic of the discourse (Sanford & Garrod, 1981). Yet others propose thatmany different factors interact to determine what is in focus (McKoon,Greene, & Ratcliff, 1993; Marslen-Wilson et al., 1993). In this paper, weare primarily concerned with semantic focusing.

Semantic focusing assumes that both verbs and connectives have focusingproperties. The focusing properties of the verb direct attention to theendpoint or consequence of the described event. The focusing properties ofthe connective depend upon its meaning. Connectives such as because

direct attention to the cause of the previously described event, connectivessuch as so direct attention to the consequences of the event. Thus, in asentence such as, John criticised Bill so he tried to correct the fault, verbfocusing highlights Bill, since Bill is the person associated with the endpointof the event of criticising. The connective, so, directs attention to theconsequences, and hence reinforces the focus on Bill. The preferredinterpretation of the pronoun he, therefore, is that it refers to Bill. Nowconsider the case where the connective is because, as in John criticised Billbecause he failed to correct his faults. In this second example, the verbcriticise once again brings Bill into focus, but now the connective directsattention to the cause of the event. Thus the effect of the connective is toshift attention away from the consequence and towards the cause. Hencethe pronoun he is less likely to be interpreted as referring to Bill. Semanticfocusing, therefore, sees the focus of an utterance as changing dynamicallyas new input is encountered (Stevenson, 1996; Stevenson, Crawley, &

Kleinman, 1994; Stevenson & Urbanowicz, submitted).


4/38


Early work on the above kind of focusing attributed all the effects to theverb (e.g., Garvey & Caramazza, 1974). According to Garvey andCaramazza (1974), the implicit causality of a verb determines who is thecausal instigator of the described event and the causal instigator is the

preferred referent of a subsequent pronoun. This emphasis on the verb andneglect of the role of the connective was a result of only using materialscontaining the connective because. Once studies included other connec-tives in addition to because, the importance of the connective becameapparent (Au, 1986; Ehrlich, 1980; Stevenson et al., 1994). Stevenson et al.(1994) examined three kinds of verbs (transfer, actions and states) and fourconnectives (so, because, and and a full stop (the null connective)). Theprotagonists in an event described by an action verb (e.g., John hit Bill /Bill was hit by John) ll Agent (e.g., John) and Patient (Bill) thematicroles. The protagonists in an event described by a transfer (e.g., John gavethe book to Bill / Bill took the book from John) ll Goal (Bill) and Source(John) thematic roles. The protagonists in a state, described by a stateverb, (e.g., John liked Bill / Bill pleased John) ll Experiencer (John) andStimulus (Bill) thematic roles. Stevenson et al. (1994) used sentencecontinuation tasks in which participants were presented with sentence

fragments ending in a connective followed by a pronoun (e.g., Johncriticised Bill and he . . .). Participants wrote continuations to the fragmentsand the continuations were examined to determine how the pronouns hadbeen interpreted.

Stevenson et al.s results are summarised in Table 1. In the table, it can beseen that the Goal and the Patient are always preferred in transfer andaction sentences, whereas the preferred thematic role in state sentencesdepends on the connective: the Experiencer is preferred with and andso,the

Stimulus is preferred with a full stop and because. It can also be seen that in

TABLE 1

Summary of the Results of Stevenson et al.s (1994) Sentence Continuation

Studies Showing which Thematic Roles were Preferred as the Referents for Pronouns

Type of verb

Transfer Action State

Type ofe.g., John passedthe comic to Bill % e.g., Joseph hitPatrick % e.g., Ken impressedGeoff %

Connective Preferred role Preferred role Preferred role

And Goal (Bill) 78 Patient (Patrick) 78 Experiencer (Geoff) 76So Goal 77 Patient 80 Experiencer 89

Full Stop Goal 67 Patient 59 Stimulus (Ken) 73Because Goal 57 Patient 64 Stimulus 87

Note: %=

percentage of pronouns referring to the preferred thematic role.


5/38


transfer and action sentences, the preferences for Goal and Patient arereduced (but not eliminated) with the full stop and because, whereas in statesentences the preference for the Stimulus with the full stop and becauseeliminates the preference for the Experiencer found with and and so.

Stevenson et al. (1994) explained these results by proposing that whenpeople encounter an event verb, they construct a tripartite mentalrepresentation of the action. This representation consists of a pre-condition (which may be the cause), the action itself, and the endpoint(which may be a consequence) of the action (Moens & Steedman, 1988).Stevenson et al. claim that the default focus in clauses describing events ison the thematic role associated with the endpoint of the event, a focus thatis attenuated when the connective directs attention to the cause. Thematicroles associated with the endpoint of the described event are Patient inaction sentences and Goal in transfer sentences. On the other hand, a statehas no tripartite representation since it has no pre-condition and noendpoint (Moens & Steedman, 1988). According to Stevenson et al. (1994),therefore, there is no default focus in state sentences. A preferred focusonly appears when a subsequent connective converts the state into an eventhaving a pre-condition and an endpoint. If the connective directs attention

to the pre-condition, as with because or a full stop (an implicit causalconnective), then the Stimulus is preferred. If the connective directsattention to the consequences, as with and or so, then the Experiencer ispreferred. These proposals are supported by both the sentence continua-tion studies (Stevenson et al., 1994) and reading time studies (Stevenson &Urbanowicz, 1995, submitted).

The relational hypothesis

An alternative candidate account of pronoun resolution is one thatemphasises the relation between eventualities

1rather than the focusing of

a particular individual participating in an eventuality (Hobbs, 1979). Thisalternative account exploits a theory of discourse structure that attributesthe coherence of a discourse to the relations that hold between differentparts of the text. There have been several such theories; see Hobbs (1985)

and Mann and Thompson (1988) for two inuential ones. These theoriescharacterise the structure of discourse in terms of relations that holdbetween the eventualities described by adjacent spans of text. Suchrelations include the CAUSE relation, the RESULT relation, theNARRATIVE relation and so on. As we saw in the discussion of focusing,

1 Following Bach (1986) we take the term event to include actions (including transfers), and

the term eventualityto cover both states and events.


6/38


Stevenson et al. (1994) have already identied a role for connectives inpronoun resolution. However, they emphasised the attention directingfunction of connectives, which contributes to the focusing of discourseentities; they did not consider the function of connectives that species the

coherence relation2

between two described eventualities.According to the relational view, the referent of the pronoun is

determined by the choice of coherence relation and not by what is infocus. For example, since a RESULT relation concerns the person wholls the thematic role associated with the endpoint of an eventuality, it isthe choice of a RESULT continuation that leads participants to interpretthe pronoun in the fragment as referring to the individual associated withthe endpoint. Our initial motivation for the relational hypothesis as analternative to the focusing hypothesis for an understanding of pronouncomprehension, came from the observation that two of the connectivesused by Stevenson et al. (1994) were ambiguous. These connectives were

so and and, which each admit two possible coherence relations. Wetherefore wished to examine the proposition that the two possibleinterpretations of each ambiguous connective coincided with twodifferent interpretations of the pronoun.

According to the taxonomy of connectives given in Knott (1996), theambiguity of so arises because so can signal either a RESULT relation,where one event is described as the result of another (see Example 1), or aPURPOSE relation, where an intended result is characterised as anAgents rationale for acting (see Example 2). And so is specic to the rsttype of relation, while so that is specic to the second.3 (The PURPOSErelation in (2) is faintly possible with and so as well as with so that, but itseems to trigger a slightly different interpretation of the relation than so

that.) Note that in (1), the pronominal referent is Bob, whereas in (2) it isBill. Our relational hypothesis is that each coherence relation will beassociated with a different pronominal referent, RESULT relationsfavouring the thematic role associated with the consequences, thePURPOSE relation favouring the Agent of an event. (Hence states,which have no Agent, should have no PURPOSE continuations.) We testthis proposition in Study 1.

{ so }(1) Bill handed the plate to Bob { and so } he (Bob) balanced his

glass on it.{ #so that }

2 Following Hobbs, we will use the term coherence relations to refer to what others havecalled rhetorical relations or discourse relations.

3 Note that the hash sign in these examples does not indicate ungrammaticality or even

incoherence, but just that the connective indicated is not suitable as a substitute for the top

phrase in the bracket. See Knott and Dale (1994), Knott and Mellish (1996) for a more preciseformulation of the notion of connective substitution.


7/38


{ so }(2) Bill handed the plate to Bob { #and so } he (Bill) could rell his

glass.{ so that }

And can signal many different relations; among them, RESULT, whichcan be made explicit with whereupon (see Example 3) and NARRATIVE,which can be made explicit with next (see Example 4). Note that thougheach phrase can be substituted by and, they are not very suitable assubstitutes for each other. In other words, they seem to be making explicitslightly different relations. Once again, our two examples show that eachpronominal referent can be associated with a different coherence relation,

the Patient with RESULT and the Agent with NARRATIVE. Weexamine these proposed relations between pronominal reference andcoherence relation in Studies 2 and 3.

{ and }(3) Bill called Bob a liar { whereupon } he (Bob) challenged him to

a duel.{ #next }

{ and }(4) Bill called Bob a liar { #whereupon } he (Bill) accused him of

cheating.4

{ next }

Relationship to other research

One way of thinking about the difference between our two hypotheses is to

say that the focusing hypothesis concerns the status of entities in thediscourse whereas the relational hypothesis concerns events and theirinter-relationships. A number of other researchers have discussed thedistinction between entities and eventualities or relations. Grosz andSidner (1986) distinguish between a theory of local focus, known ascentering theory, which is a theory of the prominence of the entities inindividual utterances, and a theory of global structure, which is a theoryconcerning the discourse purposes underlying the structure of thediscourse as a whole. Local focus is concerned with the coreferentialrelationships between utterances and is said to be responsible for theinterpretation of pronouns. Global focus is concerned with the intentionalrelations between events and is said to be responsible for the interpretation

4 It should be borne in mind that although the additional connectives introduced in this

section do seem to be more specic than those on which the experiments were carried out,they are not necessarily unambiguous themselves.


8/38


of denite descriptions. Sanford and Garrod (1981; Garrod, Freudenthal,& Boyle, 1994) distinguish between entities and events in their scenario-mapping model, although they do not explicitly discuss the relationsbetween events. According to Sanford and Garrod, entities are ranked

according to their accessibility to a subsequent pronoun, whereas the rolesthey occupy in the events described by the discourse are available asreferents for denite descriptions.

Our approach differs from the work of Grosz and her colleagues in threemain respects. First, whereas Grosz et al. (1983, 1995) concentrate onstructural focusing in their centering theory, we concentrate on semanticfocusing. Second, Grosz and Sidners (1986) global focus concentrates onthe intentional structure of the discourse in terms of a task to beaccomplished. They thus characterise the relations between discoursesegments in terms of the structural relations of dominance, andsatisfaction-precedence. By contrast, we concentrate on the rhetoricalstructure of the discourse and describe the relations between events interms of more, and ner-grained, categories. Finally, Grosz and Sidnersuggest that pronouns of the kind that we study in this paper are resolvedin local focus rather than global focus, that is, pronoun resolution is

affected by structural focusing but not by the intentional structure of thediscourse. Although this view is consistent with the focusing hypothesis, itis not consistent with the relational hypothesis.

Our approach also differs from Sanford and Garrods work in threemain respects. First, they emphasise the focusing function of generalbackground knowledge whereas we concentrate on semantic focusing.Second, Sanford and Garrods notion of role is dened by the use ofdenite descriptions (e.g., an individual may ll the role of waiter or

customer etc.) as opposed to referring to the same individual by a propername; in contrast, we believe that thematic role is the relevant notion ofrole since thematic roles explicitly encode the semantic role of an entity inan event but a denite description may not. These thematic roles aresusceptible to semantic focusing. Third, we specically propose thatpronoun comprehension is also affected by the relations between theevents described in the discourse, an aspect of discourse that is notconsidered by Sanford and Garrod.

The present study

In order to see if the relational hypothesis is a feasible alternativeinterpretation of Stevenson et al.s (1994) data, we used data from sentencecontinuation studies, either those we conducted ourselves or thoseconducted by Stevenson et al. (1994). In these studies, a subject is

presented with sentence fragments like the following:


9/38


John liked Bill and he . . .

and asked to write a continuation to the fragment. The continuations arethen scrutinised to determine who the pronoun refers to. In the presentstudy we also classied each continuation according to its relationship tothe event described in the fragment. This meant that we could determinewhether or not the interpretation of the pronoun was consistent with thecoherence relation between the two clauses. A consistent response wouldbe one in which the RESULT relation co-occurred with an interpretationof the pronoun as referring to the thematic role associated with theendpoint of the described eventuality. An inconsistent response would beone in which the RESULT relation co-occurred with an interpretation of

the pronoun as referring to the thematic role associated with the pre-condition of the eventuality.

In a sentence continuation task, the subject is engaged in twooverlapping processes: a comprehension process, involving the interpreta-tion of the fragment, including the pronoun; and a production process,involving the choice of coherence relation and its expression in thecontinuation. Thus, although the task is to produce a continuation, itrequires participants to understand the fragment before doing so. In

particular, participants responses are made on the basis of the mentalrepresentations they have developed while reading the fragment. In thispaper, we are specically concerned with the processes of comprehensionrather than production, since we aim to discover which hypothesis bestexplains how the pronoun at the end of the fragment is interpreted. Thecontinuation itself is used to inform us about that interpretation.

According to the focusing hypothesis, the mental model of the fragment

contains two entities that differ in their accessibility, with the degree offocusing depending on whether the initial focus on the endpoint of therepresented event is reinforced or reduced by the connective. If there is astrongly focused entity, as is the case when the connective reinforces thefocus on the endpoint, then the pronoun in the fragment is most likely to beinterpreted as referring to this entity. If there is no strongly focused entity,as is the case when the connective shifts attention towards the cause, thenthe likelihood of the pronoun being interpreted as referring to the entity

associated with the endpoint will be reduced. According to the relationalhypothesis, the mental model of the fragment represents an event that willhave a specied relation to the event described in the continuation. Thenature of this relation depends on how the connective is interpreted and theinterpreted relation then determines the interpretation of the pronoun. Forexample, if the interpretation of the connective favours a RESULTrelation, then the thematic role associated with the consequences of theevent described in the initial fragment will be the preferred pronominal


10/38


referent, whereas if the interpretation favours a NARRATIVE relation,the Agent will be the preferred pronominal referent.

It is likely, however, that in natural discourse the most focused referentis also the referent of the pronoun as determined by the relational

hypothesis and that the two predictions are more closely related than wasimplied in the descriptions above. This possibility poses problems fordistinguishing empirically between the two hypotheses. The basic strategywe followed, therefore, was to assume the focusing hypothesis was true andthen look to see whether or not the referent of the pronoun was alsocompatible with the relational hypothesis. If it was not, then we wouldhave evidence against the relational view and in favour of the focusingview. We followed this strategy in our rst two studies. However, in ourthird study, we used the connective, next, for which we were able togenerate differential predictions for the two hypotheses.

STUDY 1

In this study, we re-ran Stevenson et al.s Experiment 3, in which theconnectivesbecause and so were used. In their experiment, Stevenson et al.

did not include a pronoun at the end of the fragment, since they wereprimarily interested in the focusing properties of thematic roles rather thanthe role of focusing in pronoun comprehension. However, in this paper, wewish to examine how the pronoun is interpreted, hence we re-ran theexperiment, but included the pronoun at the end of each sentencefragment.

The experiment included sentence fragments containing both becauseand so; however, we only examined the continuations to so. We expect the

coherence relations expressed in these continuations will be eitherRESULTs or PURPOSEs, in agreement with the two possible meaningsof so. The focusing hypothesis predicts that the pronouns will beconsistently interpreted as referring to the thematic role associated withthe endpoint of the eventuality described in the fragment. The relationalhypothesis predicts that when the continuations are RESULTs, thepronoun will refer to the person affected by the eventuality, that is, tothe Patient. But when the continuations contain PURPOSEs, the pronounwill refer to the Agent of both transfers5 and actions. There should be noPURPOSEs in state continuations because states do not have Agents.

5 The Agent in a transfer sentence can be either the Goal or the Source, depending onwhich is the subject of the sentence. There are problems, therefore, with this analysis becauseit assumes that two thematic roles are associated with the subject of transfer verbs. However,Jackendoff (1972) has argued that an NP in a single sentence can bear more than one thematicrole (see also Cowper, 1992) and our results and those of Stevenson et al. (1994) are hard to

explain in the absence of such a claim.


11/38


However, we fully expect that Stevenson et al.s (1994) ndings will bereplicated and that the thematic role associated with the endpoint of theeventuality will be preferred pronominal referent. Hence, there should bevery few continuations expressing PURPOSEs. The relational hypothesis,

therefore, predicts that the continuations should predominantly expressRESULTs, consistent with the pronoun referring to the thematic roleaffected by the eventuality.

Method

Participants. Thirty-two undergraduate volunteers served as partici-pants and their ages ranged from 18 to 32.

Design and materials. The connectives so or because were used toconnect a clause containing a pronoun to a clause introducing twoindividuals who were both potential antecedents for the pronoun. Forexample, Ken admired Geoff so/because he . . . . The materials were thesame as those used in Stevenson et al. (1994). There were three kinds ofsentences: those containing transfer verbs with Goal and Source thematic

roles; those containing action verbs with Agent and Patient thematic roles;and those containing state verbs with Experiencer and Stimulus thematicroles. Denitions of the roles used by Stevenson et al. are given in Table 2.The denitions in the table were gleaned from Andrews (1985), Fillmore(1968), Jackendoff (1985) and Radford (1988). Each sentence occurred intwo versions, one in which one thematic role was mentioned rst, the otherin which the alternative thematic role was mentioned rst. An example ofeach kind of sentence and each version is shown in Table 3. The

participants wrote continuations to 48 sentence fragments, 16 containingeach verb type. Hence for each verb type, there were four fragments ineach of the four conditions dened by sentence version and connective.

TABLE 2

Denitions of Thematic Roles used in Stevenson et al. (1994)

a. Goal: someone or something towards which something moves. Examples: Mary in Johngave the the book to Mary. Peter in Peter took the book from Susan.

b. Source: someone or something from which something moves. Examples: John in Johngave the book to Mary. Susan in Peter took the book from Susan.

c. Agent: the instigator of an action. Examples: subjects of smash, kick, criticise, reproach.d. Patient: someone or something affected by an action. Examples: objects of kill, eat,

smash, but not those of watch, hear, and love.e. Experiencer: someone or something having a given experience. Examples: subject of

love, object ofannoy.f. Stimulus: someone or something giving rise to a certain experience. Examples: object of

love, subject of annoy.


12/38


Categorising the continuations. The continuations were categorisedaccording to the coherence relations they expressed. The basic procedurefor categorising the continuations was the same for all the data examinedin this paper. In all three studies, two judges categorised half each of thecontinuations produced for one sentence type. The judges were all

undergraduate volunteers who were paid for their services and who wereblind to the experimental hypothesis and theoretical framework. Beforestarting the categorisations, all the judges were given the informationshown in Table 4 and had the categories explained to them. It was alsopointed out to them that some continuations could be interpreted as eitherRESULTs or NARRATIVEs. For example.

Colin threw the ball to Gary and he dropped it as usual.

The judges were instructed that, in these circumstances, they should assignthe continuations to the RESULT category. RESULTs were chosen totake precedence over NARRATIVEs because it was thought thatRESULTs expressed a stronger relation between the two events thandid NARRATIVEs and so judges were instructed to opt for the stronger ofthe two relations in these ambiguous cases. In an initial reliability check,the two judges rst categorised the continuations from the same threeparticipants after which they checked their degree of agreement anddiscussed any disagreements until they reached a consensus view. The judges were told that if they disagreed on 20% or more of thecontinuations, they were to seek advice from the rst author. Thissituation did not arise in any of the categorisations in the study. Afteragreeing the initial disagreements, each judge then categorised half of theremaining continuations. At the end of this categorisation task, each judgethen categorised the continuations of three participants selected at random

(but not including those used in the initial reliability check) from those

TABLE 3

Examples of Sentence Fragments Used in the So/BecauseStudy

(Experiment 3 of Stevenson et al. 1994)

Version one Version two

Transfer Sentences Goal-Source Source-Goal John seized the comic from Bill John passed the comic to Billso/because . . . so/because . . .

Action Sentences Agent-Patient Patient-Agent Joseph hit Patrick Patrick was hit by Joseph

so/because . . . so/because . . .

State Sentences Experiencer-Stimulus Stimulus-Experiencer

Ken admired Geoff Ken impressed Geoff

so/because. . .

so/because. . .


13/38

TABLE4

CriteriaforAssigningCoherenceR

elationsintheCompletionstoCategoriesandE

xamplesofEachCatego

ry

Categoryof

continuation

DescriptionofEachCategory

Criterionfordec

idingthecategoryofthe

continuation

Examples

Result

Theeventdescribedinthecontin

uation

happened

asa

consequenceof

what

happenedintherstclause.

Thewholesentencemakessensewhen

theconnectiveisreplacedwithsoor

withasaresult.

Sometimesthewholesentencealsots

thecriterionfor

theNarrativeCategory.

AlwaysgivepreferencetotheConse-

quencecategory.

Johnseizedthecomicfrom

Billandhe

triedtoseizeitback.

KenadmiredGeoffandhegavehim

a

job.

JosephhitPatrickandh

eranoffcrying.

Narrative

Theeven

tdescribedinthecontinua

tionis

one

thatjusthappened

nextw

ithout

beinga

consequenceoftheeventde-

scribedintherstclause.

Thecontinuationmakessensewhenthe

connectiveisrep

lacedbyThenorAfter

that,

butdoesn

otmakesensewhenthe

connectiveisreplacedbySoorAsa

result.

Johnseizedthecomicfrom

Billandhe

begantoreadit.

CliveangeredFredand

heignoredhim.

TimothyhelpedIanan

dhethenwent

inside.

Background


uation

happenedatthesametimeasthe

event

described

in

the

rstclauseo

r

the

continuation

givesthecontextofthe

eventdescribed

in

therstclau

seor

sentence.

Noobviousconn

ectivetoreplacetheone

inthecontinuation.

Malcolm

gavesomemoneytoStewart

andheknewhewantedtospenditon

booze.

AndrewshockedJeremyandheknewit.

JosephhitPatrickand

hekepthitting

him.

Purpose


uation

describeswhatisnowpossiblegiventhe

eventde

scribedintherstclause.The

eventde

scribedinthecontinuatio

nmay

nothave

takenplaceyet.

The

continuatio

n

usually

containsso

thatandthev

erbcontainscouldor

might.

Julierolledtheballto

Rachelsoshe

couldscore.

DickdeceivedCarlsothathecouldgain

theupperhand.

JosephhitPatricksohecouldbeexpelled

fromschool.

237


14/38


categorised by the other judge. The degree of agreement between the twojudges was then calculated to give an overall reliability measure. In thisstudy, the two judges agreed in 93% of the cases.

Results

We rst checked that the pronoun interpretation results were the same asthose obtained by Stevenson et al. (1994). This turned out to be the case.When the sentence fragments contained transfer or action verbs, thepreferences for Goal and Patient were reinforced with so and reduced bybecause; when the fragments contained state verbs, there was a preferencefor the Experiencer with so and for the Stimulus with because.

The percentage of PURPOSE and RESULT continuations to sofragments in which the pronoun referred to the preferred thematic roleare shown in Table 5. Completions in which the pronoun referred to thenon-preferred thematic role are not shown because the numbers were verysmall.

Since the frequency of continuation in a category was not independentof the frequencies in other categories, it was not possible to compare thefrequencies of the different categories within each sentence type. Instead,

TABLE 5

Percentage of Result and Purpose Relations in the So Completions of Study 1 where

the Pronoun Referred to the Preferred Thematic role

Type of completion

Type of verb Sentence version Result Purpose

Transfer Goal-Source 19 42 John seized the comic from Bill so he read it. could read it.

Source-Goal 22 54 John passed the comic to Bill so he read it. could read it.

Action Patient-Agent 51 06

Patrick was hit by Joseph so he cried. could show howbrave he was.

Agent-Patient 56 10

Joseph hit Patrick so he cried. could show how

brave he was.State Exper.-Stimulus 69 0.0

Ken admired Geoff so he gave him the prize. Stimulus-Exper. 65 02

Ken impressed Geoff so he gave him the prize. became very bigheaded.

Notes: The preferred thematic roles are in bold; the most frequent continuations are also inbold. Example fragments are shown in the Sentence Version column. Examples of

completions are shown in the Result and Purpose columns.


15/38


we compared the results to chance by conducting one sample t-tests on thecontinuations that referred to the preferred thematic roles; the results weretoo sparse for continuations referring to the non-preferred thematic rolesto be analysed. Because four comparisons were made on each verb type

(two versions two continuation types), we adopted an alpha level of .01for each comparison. Two analyses were carried out on each comparison,in accordance with Clarks (1973) procedure, one treating participants as arandom effect and the other treating sentences as a random effect. Therewere 31 DF in the participants analyses and 15 DF in the sentencesanalyses.

It is difcult to determine the chance frequency of each type ofcontinuation because the number of possible categories is potentially verylarge. However, we decided to base our estimate on the two categories(PURPOSE and RESULT) that we had identied before we startedclassifying them and to allow for a third catchall category for othercontinuations. Since there were two possible referents for the pronoun andthree types of continuations (RESULT, PURPOSE, other), there was onechance in six (i.e., a 17% chance) of a category being produced in eachcondition.

Transfer verbs: Pronoun refers to Goal. The number of PURPOSEcontinuations was signicantly greater than chance in both Goal-Sourcesentences (participants: t = 4.78, p < .001; sentences: t = 3.25, p < .004)and Source-Goal sentences (participants: t= 4.78, p < .001; sentences: t=15.47, p < .001). RESULT continuations did not differ from chance ineither Goal-Source (participants: t< 1; sentences: t < 1) or Source-Goalsentences (participants: t < 1; sentences: t = 1.15).

Action verbs: Pronoun refers to Patient. As can be seen in Table 5,RESULT continuations predominated with action verbs. The number ofthese continuations was signicantly greater than chance in both Agent-Patient (participants: t= 6.65, p < .001; sentences: t= 6.32, p < .001) andPatient-Agent sentences (participants: t = 6.54, p < .001; sentences: t =4.85, p < .001). PURPOSE continuations were signicantly less frequentthan chance in Patient-Agent sentences (participants: t = 6.08, p < .001;sentences: t = 4.0, p < .002), whereas they did not differ from chance inAgent-Patient sentences (participants: t = 2.24, p < .04; sentences: t =1.84, ns).

State verbs: Pronoun refers to Experiencer. Inspection of Table 5indicates that RESULT continuations predominated. The numbers ofthese continuations were signicantly greater than chance in both

Experiencer-Stimulus (participants: t = 10.33, p < .001; sentences: t =


16/38


10.94, p < .001) and Stimulus-Experiencer sentences (participants: t =8.59, p < .001; sentences: t = 8.27, p < .001). The number of PURPOSEcontinuations was signicantly lower than chance in Stimulus-Experiencersentences (participants: t = 8.36, p < .001; sentences: t = 4.66, p < .001).

There were no PURPOSE continuations in Experiencer-Stimulus sen-tences.

DISCUSSION

According to the focusing view, the choice of a pronomial referent inaction and transfer sentences is a function of a default focus on thethematic role associated with the endpoint of the event, a focus that ismaintained when the connective is so. In state sentences, the connectivemust be encountered before attention can be directed to the endpoint ofthe state. The results in this study were, as expected, consistent with thishypothesis. The preferred pronominal referent was Goal in transfers,Patient in actions, and Experiencer in states. The question we now need toask is whether or not these same results could also have been predicted bythe choice of coherence relation in the continuation. That is, do these

preferred pronominal interpretations co-occur with RESULT coherencerelations?The answer to this question is afrmative for actions and states, but not

for transfers, where PURPOSE was the predominant coherence relation.Thus, our prediction that RESULTs would predominate with all threeverb types was not conrmed. A strong view of the relational hypotheses,therefore, cannot be maintained in the light of our results. They indicatethat PURPOSEs are not restricted to Agents and so suggest a weak view

of the hypothesis, in which a unique coherence relation is consistent withthe interpretation of the pronoun within each verb type, but notnecessarily across verb types. The present results t this weak view: thechoice of PURPOSE relations in transfer continuations was accompaniedby Goal pronominal referents; the choice of RESULT relations in actioncontinuations was associated with Patient pronominal referents; and thechoice of RESULT relations in state continuations was accompanied byExperiencer pronominal referents.

What, then, might explain the choice of PURPOSE continuations intransfer sentences rather than RESULTs? One possibility is that transferverbs have a different semantic structure from actions and states. Whereasan action or state can be the cause of a subsequent eventuality (e.g., hittingsomeone causes them to cry; hating someone causes you to want to hurtthem), a transfer does not seem to cause another eventuality. Rather itseems to create a condition that enables the Goal to do something

(Goldman, 1986). For example, passing a book to Bill does not cause him


17/38


to read it, but it does enable him to read it. Consequently, a RESULTrelation, as in the continuation, he read it, in John gave the book to Bill sohe read it is less likely to be produced because it presupposes a causalstructure underlying the transfer whereas there is only an enablement

structure. Hence PURPOSEs, which do not conict with the enablementstructure, are preferred. Indeed, it could be argued further that theconsequences of a transfer are part of the meaning of the verb itself, theseconsequences being that the Source no longer possesses the object and thatthe Goal now possesses the object. Our participants, therefore, producedcompleted sentences like John passed the comic to Bill so he could read it,because the initial fragment ( John passed the comic to Bill so . . .)presupposes the consequence that Bill (the Goal) now possesses the comic,and it is this consequence that enables the intended action described in thecontinuation (Bills reading the comic) to be carried out.

Overall, however, the critical nding is that in all three sentence types,thematic role focusing and coherence relations go together. In action andstate continuations, the pronoun refers to the focused thematic roleassociated with the consequence of the eventuality, and the coherencerelation is one of RESULT, in which the consequence for the individual is

described. In transfer sentences, the pronoun also refers to the thematicrole associated with the endpoint of the event but here the coherencerelation is one of PURPOSE. Thus, the ambiguity of so allowed theparticipants to select the meaning of the connective that maintainedconsistency between the meaning of the verb, the focused entity, thecoherence relation and the interpretation of the pronoun. In summary,therefore, our results are consistent with both hypotheses, and furthersuggest that language users strive to keep verb meaning, focusing,

coherence relation and pronominal interpretation in alignment.

STUDY 2

In the Introduction, we pointed out that the connective and is ambiguousbetween a meaning characterised by and then and a meaning characterisedby whereupon. The former meaning is associated with a NARRATIVErelation, whereas the latter is associated with a RESULT relation. Hence,when we consider continuations to and rather than so, as we do in thissecond study, we are concerned with the NARRATIVE rather than thePURPOSE relation. In this study, therefore, we re-analysed the andcontinuations from Stevenson et al.s Experiment 1 to see if they revealedits ambiguity between NARRATIVE and RESULT coherence relations,and, if they did, whether a specic interpretation of the connective co-occurred with the interpretation of the pronoun. Consistent with the

focusing hypothesis, Stevenson et al.s Experiment 1 results showed that


18/38


the preferred referent of the pronoun was the thematic role associated withthe endpoint of the eventuality. According to the relational hypothesis, wewould expect this preferred pronominal referent to be associated withRESULT continuations, at least with actions and states. If, however, both

RESULTs and NARRATIVEs are found with the same pronominalreferent, then this would be evidence against the relational hypothesis.

With the transfer sentences, we sought to conrm the ndings of theStudy 1. On the basis of those ndings and our interpretation of them, wewould expect transfer continuations to be mainly NARRATIVEs ratherthan RESULTs, since we have proposed that transfers already encode theresult of the action in the meaning of the verb. In other words, we expectthat, wherever possible, participants will endeavour to interpret theconnective in a way that maintains consistency between the meaning of theverb, the focused entity, the coherence relation and the interpretation ofthe pronoun.

Method

This experiment used 32 participants who each completed 64 sentencefragments. Sixteen fragments contained transfer verbs, 16 contained action

verbs, 16 contained state verbs, and 16 were in a control condition thatcontained motion verbs. The control condition is not re-analysed here.There were, therefore eight sentences in each sentence version for eachverb type. The procedure for categorising the continuations was the sameas in Experiment 1. Six judges were used, two for each verb type, eachjudge categorising half the continuations, together with the continuationsof six additional participants in order to do the initial and the nal

reliability checks. The nal reliability check showed 90% agreement.

Results

The categorisations revealed that the two predicted relations predomi-nated. However, a third relation, that of BACKGROUND, also appearedin sufcient numbers to be included in the analyses of the results. Table 6shows the mean number of each of these three categories of continuationwhen the pronoun referred to the preferred thematic role.

One sample t-tests were carried out on the continuations that referred tothe preferred thematic roles. As was the case in the previous re-analysis,the results were too sparse to do any statistical analyses when thecontinuations referred to the non-preferred thematic role. We estimatedchance level on the basis of the two possible pronoun interpretations andfour possible continuation categories: RESULT, NARRATIVE, BACK-GROUND, and OTHER. Thus there was a 1 in 8 (i.e. 12.5%) chance of

producing a given continuation. Since six statistical comparisons were


19/38


made for each verb type (two sentence versions three continuationtypesthe OTHER category was not analysed), the alpha level was set at.008. There were 31 DF in the participants analyses and 15 DF in thesentences analyses.

Transfer verbs: Pronoun refers to Goal. NARRATIVE continuationswere the most frequent and were produced signicantly more often thanchance both when Goal was mentioned rst (participants: t = 10.63, p

Date post:	30-May-2018
Category:	Documents
Upload:	reypaps
View:	216 times
Download:	0 times

stevensonetal2000

Documents